DavidAU/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated
DavidAU/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated is a 4 billion parameter Qwen3-based language model, uncensored and abliterated using the Heretic method. It features a 40960 token context length and is specifically optimized for unrestricted content generation with a low refusal rate (8/100) and minimal KL divergence (0.06), ensuring model integrity. This model is designed for diverse use cases requiring honest, uncensored responses without judgment.
Loading preview...
Model Overview
DavidAU/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated is a 4 billion parameter model based on the Qwen3 architecture, distinguished by its "abliterated" and uncensored nature. This model has undergone a de-censoring process using the Heretic v1.0.1 method, resulting in a significantly reduced refusal rate of 8/100 compared to the original model's 87/100. A key focus during its development was maintaining model integrity, evidenced by a very low KL divergence of 0.06, indicating minimal "brain damage" from the abliteration process.
Key Characteristics
- Uncensored Output: Designed to provide honest, unfiltered responses across all topics without refusal.
- High Context Length: Supports a substantial context window of 40960 tokens.
- Low Refusal Rate: Achieves an 8/100 refusal rate, making it highly compliant with user requests.
- Model Integrity: A KL divergence of 0.06 ensures the model's core capabilities are preserved post-abliteration.
Usage Considerations
While uncensored, this model may require explicit direction or "pushing" with specific slang or terms to generate content at desired graphic or explicit levels, especially for x-rated or highly descriptive content. It is part of the broader Qwen3-24B-A4B-Freedom-Thinking-Abliterated-Heretic-NEO series, emphasizing user freedom and versatility. For optimal performance, specific sampler and parameter settings, such as a Smoothing_factor of 1.5, are recommended for chat and roleplay applications.